CIENCiena Corporation

“In just the last few weeks, we have seen announcements from the four largest global hyperscalers that outlined a step-function increase in their 2026 CapEx to more than $600 billion in aggregate, driven by infrastructure needs related to AI training and inference workloads at massive scale.”

“we went from 2024 to 2025, that doubling sort of in the rearview mirror, and then we talked about it as a major portion of our inside and around the data center with our aspiration to triple that this year, and we are well on track for that.”

“We have actually seen during the course of this quarter additional sites being added to that. Again, I would say all of these currently that we are seeing are in North America, which is, I think, to be expected. We have added two more hyperscalers to that that are also rolling this out. I think we are in the very, very early stages of this, and in talking with them, though, the plans are large and expansive, as you would expect for the scale of what they are trying to do here.”

“For scale-up opportunities inside the rack, where XPUs are getting faster and driving heat and power concerns, we are advancing the Nitro Linear Redriver technology, also from our Nubis acquisition. We believe this is a critical element to active copper cabling solutions, which extend the distance that signals can travel and reduce power by up to 80% versus AEC-type solutions. We also expect samples of the Nitro Redriver to be available in calendar Q2 2026.”

“we are in the very early stages of this wave of opportunity, and we are already experiencing extraordinary demand, with three hyperscalers choosing to use our optical solutions for their training applications across distance, which we have talked to you about in recent quarters. And all three hyperscalers are significantly ramping, including additional orders for multiple additional from the first hyperscaler we announced in Q3 2025.”

“our data center out-of-band management, or DCOM, solution continues to represent another significant opportunity inside the data center. Leveraging our XGS-PON and routing and switching platforms, DCOM was initially designed with Meta to meet hyperscale provisioning and configuration requirements. We continue to work with them and are engaged in technical discussions with two other major global hyperscalers.”

“Last week, we announced the Vesta 206.4T optical engine, which is the industry's first high-density, low-power, open-ecosystem pluggable CPO solution. Samples of the Vesta product will be available in calendar Q2 2026, and we are actively discussing Vesta, as you would expect, with our cloud provider customers and partners, and we are excited to be showcasing it at OFC again in a few weeks' time.”

“These record results reflect Ciena Corporation's market leadership and reinforce our role as a critical provider of the high-speed optical systems and interconnects that enable AI workloads to scale and to be monetized. In fact, we are taking meaningful share of the increase in AI-driven connectivity spend as customers trust our technology leadership, deep collaboration, and proven execution.”

“We expect to expand our role in scale across applications with the introduction of our new RLS HyperRail solution. HyperRail delivers an order-of-magnitude increase in fiber density within existing rack footprints, helping customers scale traffic while reducing, and in some cases avoiding, costs and complexity associated with adding substantial numbers of amplifier huts.”

“We are addressing this demand for scale across solutions with our RLS platform, the de facto industry line-system standard for cloud providers, as well as our 800ZR pluggable optics. To underscore this, we realized a second consecutive record quarter for RLS shipments and revenue.”

“You are seeing growth in cloud, general cloud. You are seeing inference. You are seeing this new market of training now emerge. As I said in my comments, we have now got three hyperscalers deploying us for training, and we are at the very, very early stages of that.”

“This solution, we expect, will begin standardization at 2026 and will ramp in 2027, allowing us to capture share and incremental value as these distributed AI training expands across regional clusters and moves to further distances.”

“As we know, the physics of copper inside the data center is reaching its limit. While there will be a place for copper solutions with shorter distance scale-up interconnects, network architectures will include more optical co-packaged interconnects, and over time, as data rates and bandwidth requirements continue to increase, coherent optical connections will overtake IMDD ones for shorter reaches to address growing capacity volumes inside the data center.”

“It is also increasingly fueled by the rise of large-scale AI data centers that need to be interconnected with DCI solutions linking data centers across campuses, regions, and continents. Additionally, service providers around the world have begun reinvesting in their optical transport infrastructure alongside autonomous networking capabilities, both to support surging AI-driven traffic growth across their networks and to improve operating efficiencies.”

“I think there are a number of elements to that sustainability. I think it is deep collaboration, first off, and understanding and intimacy, and obviously Meta were incredibly helpful in instigating that. But there are different use cases; they are slightly different in the different hyperscalers. The dependability of it is we are very vertically integrated into it. We own the core technology, and it is the software that we are putting on that as well.”

“There has been such a maniacal focus—and continues to be, obviously—on things like power, GPU accessibility, etc. Now it is really about the network. The traffic is beginning to come out of the network both for inference and for training. And the neoscalers are obviously seeing that too.”

“We see multiple waves of opportunity ahead, from continued AI training to expanding inference workloads, both domestically and internationally, to new HyperRail solutions and faster interconnects inside the data center as higher-speed requirements come online.”

“Furthermore, to monetize AI for both training and inference workloads—the latter of which represents another significant growth vector still in its infancy—the foundational requirement is again high-speed connectivity.”